Submitted to the 47th AIAA Aerospace Sciences Meeting, Jan 2009

Calibrating the γ-Reθ Transition Model for Commercial CFD

Paul Malan1
CD-adapco, Lebanon, NH 03766

Keerati Suluksna2 and Ekachai Juntasaro3

Suranaree University of Technology, Nakhon Ratchasima 30000, Thailand

The correlation-based γ-Reθ transition model is designed for implementation in modern

unstructured, parallel CFD codes. So far, this model has been published in incomplete form,
with two key correlations, Reθc and Flength, being kept proprietary. This paper describes the
implementation of the γ-Reθ transition model in a commercial CFD code, including a
description of the process of calibrating the key correlations using published data. Results
are presented for flat plate boundary layers, with and without pressure gradient, and for
airfoil and turbomachinery flows. A 3D multi-element airfoil simulation is presented as an
example of an industrial test case.

Nomenclature
k = turbulent kinetic energy
C = chord length
Cf = skin friction coefficient
Cp = pressure coefficient
Flength = function to control transition length
Fonset = function to control transition onset location
L = flat plate length
Reθ = momentum thickness Reynolds number
Reθc = momentum thickness Reynolds number where the intermittency starts to increase
Reθt = momentum thickness Reynolds number where the skin friction starts to increase
Reθt = transported variable for Reθt
S = streamwise distance
Tu = turbulence intensity expressed as a percentage
U = velocity magnitude
u* = friction velocity
x = distance along flat plate
y = wall-normal distance
y+ = non-dimensional wall-normal distance, ρyu*/ν
γ = intermittency
δ99 = boundary layer thickness
µ = dynamic viscosity
µt = turbulent viscosity
ρ = density
ω = specific dissipation rate

1
Senior Research Engineer, CD-adapco, New Hampshire Office, 21 Lafayette Str., Suite 230, Lebanon, NH 03766.
Senior Member.
2
Lecturer, School of Mechanical Engineering, Institute of Engineering, Suranaree University of Technology,
Nakhon Ratchasima 30000, Thailand.
3
Associate Professor, School of Mechanical Engineering, Institute of Engineering, Suranaree University of
Technology, Nakhon Ratchasima 30000, Thailand.

1
American Institute of Aeronautics and Astronautics
092407
 I. Introduction

T RANSITIONAL boundary layer flows are important in many CFD applications of engineering interest, such as
airfoils, wind turbines, boat hulls and turbomachinery blade rows. Unfortunately, modern unstructured, parallel
CFD codes do not lend themselves to traditional correlation-based methods for transition prediction. These methods
are based on non-local variables such as momentum thickness or boundary layer edge location, which can be
expensive, impractical, or even practically impossible to evaluate. The γ-Reθ transition model, introduced by Menter
et al. 1 in 2004, presented for the first time a correlation-based approach to transition modeling that was designed
specifically for modern CFD codes. Unfortunately, the model has not gained wide acceptance in the CFD
community because two critical correlations were deemed proprietary and have remained unpublished by the
original authors, even as their model has been refined in subsequent publications.2-5 Arguably, the model is a useful
framework into which prospective users might insert their own correlations based on their own data and/or data from
the public domain. Few CFD practitioners, however, have the resources or expertise to develop such correlations.
Recently, efforts by independent research groups to synthesize the two missing correlations have started to bear
fruit.6-7 A systematic effort at the Suranaree University of Technology, Thailand,7 has resulted in proposed forms for
these correlations that are believed to capture the essential behavior of the γ-Reθ transition model as documented by
Menter et al., 1-5 along with a viable methodology for “tuning” the correlations for a specific CFD code. Such tuning
is thought to be necessitated by small implementation differences that might exist between different codes, such as
the discretization approach or the adoption of alternative variants of the base turbulence model. These small
differences can affect the ability of the correlations, calibrated in one code, to function accurately in a different code.
In this paper, the implementation of the γ-Reθ transition model in a modern commercial CFD code, STAR-
CCM+8 is discussed and the forms of the missing correlations are described and analyzed. The calibration approach
using zero-pressure gradient, flat plate data is presented first, followed by tests of the model against non-zero
pressure gradient test cases for which data are freely available. Finally, an example of an industrial application of the
model to a 3D multi-element airfoil is presented.

II. Model Implementation

STAR-CCM+ has been developed with a client-server architecture using object-oriented methods in C++ and
Java. It uses a cell centered, finite volume discretization approach applied to cells of arbitrary polyhedral shapes.
Two flow solver algorithms are available: a segregated approach using Rhie-Chow interpolation (see, for example,
Demirdzic and Muzaferija9) and a coupled approach using the preconditioning of Weiss and Smith.10 A wide range
of turbulence models is provided, including the SST k-ω model of Menter11-12 which is a prerequisite for the γ-Reθ
transition model. STAR-CCM+ embodies the state-of-the-art of modern software development methods, which
allows rapid model implementation. Nevertheless, the approach described herein for calibrating the γ-Reθ transition
model could be applied to any CFD code.
The formulation of Menter’s SST k-ω turbulence model11-12 and the γ-Reθ transition model1-5 in STAR-CCM+ is
given in the appendix. The turbulence model is a reasonably standard implementation with several exceptions. First,
the wall boundary condition for ω is specified within the wall cell, rather than at the boundary as suggested by
Menter.11 This allows compatibility with wall functions to be maintained, although only low-Reynolds number wall
treatments are used in the present study. Second, in accordance with later recommendations,12 the strain rate tensor
modulus is used in the production term rather than the vorticity as in the original model formulation.11 Third, the
realizability constraint of Durbin13 is used as opposed to the less restrictive bound on the production favored by the
originator of the SST k-ω turbulence model.12 Experience with validating and applying Durbin’s constraint to
industrial simulations in STAR-CCM+ has thus far shown no significant adverse effects.
The transport equations for both the SST k-ω turbulence model and the γ-Reθ transition model are implemented
in STAR-CCM+ with a second-order upwind scheme. This advection scheme was also used for the momentum
equation, and where applicable, the energy equation.
A feature of STAR-CCM+ that lends itself to the implementation of the γ-Reθ transition model as a customizable
framework is User Field Functions. These are single-line, interpreted C-syntax statements that allow the
manipulation of fields (stored variables) and other User Field Functions. These may be predefined in the model, but
can also be created and easily modified by the user.

2
American Institute of Aeronautics and Astronautics
092407
 The γ-Reθ transition model requires that the transition onset momentum thickness Reynolds number, Reθt, be
specified in the free stream. Since Reθt is a non-local quantity, two noteworthy complications arise. First, one must
define what (or where) the free stream is. Second, the source term evaluation of transport equation for Reθt requires
that cells inside the boundary layer reference the value of Reθt in the free stream. In the STAR-CCM+
implementation, the first issue has been addressed
somewhat vaguely by allowing the user to specify the
location of the free stream in terms of an iso-surface
defined by User Field Function. A typical iso-value
definition might be based on wall distance (used in
most of the test cases reported herein) or the value of
the vorticity vector magnitude. The second issue is
addressed by using a KD tree algorithm to store the
location of the mesh faces that most closely
correspond to the free stream definition. The flow
variables of interest are then interpolated from the cell
values straddling the face. This introduces an
unavoidable computational overhead when run in
parallel, since the KD tree must be broadcast to each
parallel node. Since the free-stream edge can
potentially vary with each iteration, the tree cannot be
Figure 1. Comparison of Reθt correlations.
stored, so the tree is updated only every n iterations as
a cost saving measure.

III. Model Calibration Using Zero-Pressure Gradient Flat Plate Data

As described in the appendix, three correlations are required to close the γ-Reθ transition model: Reθt, Reθc and
Flength. The first of these have been presented in two alternative forms1,5 so one only has to make a choice between
them. The other two correlations are derived from calibration against zero-pressure gradient flat plate data.
Reffering to Eqs. (19) and (24), it can be seen that for the case of zero pressure gradient, λ = K = 0, so that F =
Fλ,K = 1. These two correlations are compared in Fig. 1 for zero pressure gradient. The curve of Eq. (19) is a better
fit to the experimental data cited in Refs. 1 and 5. However, according to Langtry,5 the rationale for changing this
curve was “to improve the results” for Tu < 1%. Both correlations were implemented in STAR-CCM+ for the
purpose of evaluation, but as will be explained, Eq. (24) was finally adopted.
The parameters Reθc and Flength are both expressed as functions of Reθt, the transported transition momentum
thickness Reynolds number. Viable forms for Reθc and Fonset were obtained by physical intuition and numerical
experiment as described by Suluksna et al.7 In this process, it is hypothesized that Reθc is a low-order polynomial
(preferably linear) function of Reθt. The rationale for this hypothesis is as follows. In the free stream, Reθt = Reθt by
design. One can also assume that Reθc < Reθt, since Reθc is the momentum thickness Reynolds number at which
intermittency first starts to grow, and Reθt is the momentum thickness Reynolds number where the skin friction
starts to increase. Therefore, the form Reθc = aReθt + b seems like a reasonable starting point, where a is some
number of order one but less than unity. Furthermore Langtry5 suggests a lower bound of 20 on Reθt, which will in
turn bound Reθt and Reθc. It is also clear from the model formulation that adjusting the coefficients in a manner that
increases Reθc will result in an earlier transition onset. This can be seen from the model equations, in which
increasing Reθc increases Fonset, the parameter governing where intermittency first starts to increase in the boundary
layer.
Simulations based on the ERCOFTAC data13 for cases T3A, T3B and T3AM were used to find the form of the
Flength function.7 By assuming a linear function for Reθc as suggested above, these numerical experiments showed
that Flength is a small number for large values of Reθt,(i.e., low turbulence intensities) and vice versa. Furthermore, the
solution is relatively insensitive to the exact value of Flength for small values Reθt,(i.e., high turbulence intensities).
Since the ERCOFTAC data correspond to moderately low- to high- turbulence intensity bypass transition, the
natural transition skin friction data of Schubauer and Klebanoff,13 digitized from Langtry5 and termed TSK herein,
were added to the calibration data sets.

3
American Institute of Aeronautics and Astronautics
092407
 The calibration process started with an assumed value of the coefficients a and b for the Reθc curve (say a = 0.8,
b = 0) and an assumed constant value for Flength, say (0.3 < Flength < 0.7). The lower free-stream turbulence cases TSK
and T3AM cases were then run, and the coefficients adjusted to get the best curve fit to the data. The T3A and T3B
cases could then be run to update the coefficients in an iterative process. The approximate Flength curve shape for
T3A and T3B was established by assuming that Flength is proportional to some representative wall value of Reθt, such
as the value at the leading edge. This numerical tuning resulted in the following expressions:

(
+ 62, Re
Reθ c = min 0.625Reθt


θt ) (1)

Flength = min 0.01exp −0.022 Re

(
+ 12 + 0.57,300 
) (2)
 θt


Eqs. (1) and (2) are shown graphically in

Fig. 2. The grey area represents the range of 10000
Tu>1% Tu<1%
Flength values which would be viable for high 1250
free-stream turbulence intensities. We chose
1000 Flength=min(0.1exp(-0.022Reθt+12)+0.57, 300)
the lower bound of the envelope since this
appeared to improve solution convergence. 1000
It should be mentioned that these 100
calibrated correlations are valid only when used e θc
=R
θt 750
with the Reθt correlation of Eq. (24). A similar
Flength

calibration procedure was successfully carried 10

Reθc
out for Eq. (19), but was not pursued further
500
since a quadratic polynomial form for Reθc was 1
required to reproduce the required
correspondence to the experimental data. It
0.1 Reθc=min(0.625Reθt+62), Reθt) 250
seems plausible that this is the reason behind
the modification of the Reθt correlation as
reported by Langtry.5 0.01 0
The computational mesh used for the 0 250 500 750 1000 1250
calibration studies is shown schematically in Reθt
Fig. 3. The domain extended from 0.15m
upstream of the plate to 1.7m downstream. The Figure 2. Calibrated correlations for Reθt and Flength
domain height was 0.3m. A specified constant
velocity inlet was used together with a specified
constant static pressure at outlet. Slip walls Case Uin (m/s) kin (J/kg) ωin (/s)
were used on the top boundary and upstream of
TSK 50.1 0.0827 5134
the plate. The inflow boundary conditions are
T3AM 19.8 0.0780 680.4
reported in Table 1. Fluid density was taken to
T3A 5.40 0.1386 712.2
be 1.2 kg/m3 and viscosity 1.8x10-5 PaS.
The mesh contained 100x135 quad cells, T3B 9.40 0.8964 573.2
with a near-wall cell height of 10-5m, ensuring
Table 1. Inflow boundary conditions for zero
pressure gradient flat plate simulations.

Figure 3. Mesh schematic for zero pressure gradient flat plate simulations.
4
American Institute of Aeronautics and Astronautics
092407
that the wall-adjacent cell centroid height typically ranged between 0.2 and 0.5 viscous units. Since the model was
calibrated using this mesh, particular attention was paid to mesh sensitivity. It was found that the T3AM and TSK
cases were especially sensitive to the streamwise spacing near the end of the plate, and also to the domain height.
Results obtained by doubling the mesh resolution in both coordinate directions confirmed that the mesh described
herein produces a mesh-independent solution.
The skin friction coefficient results of the TSK, T3AM, T3A, and T3B simulations are shown in Fig. 4, with the
free-stream turbulence intensity decay profiles in Fig. 5. For case TSK, the turbulence decay profiles are not readily
available, so the inflow boundary conditions were chosen to match the leading edge values of turbulence intensity
and viscosity ratio (RT) quoted by Langtry.5 It should be pointed out that these results are not evidence of the
predictive capabilities of the γ-Reθ transition model, since the model correlations were specifically adjusted to cause
the model to properly match the data. Overall, the skin friction coefficient results are comparable to those presented
by Langtry.5

Figure 4. Skin friction coefficient results for zero Figure 5. Freestream turbulence decay for zero
pressure gradient flat plate simulations after model pressure gradient flat plate simulations.
calibration.

IV. Validation Against Non-Zero-Pressure Gradient Flat Plate Data

ERCOFTAC cases T3C1-5 test the combined effect of pressure gradient and free-stream turbulence decay on
the transition prediction. They are considerably more difficult to set up since they require the prescription of a
variable free-stream velocity, as well as a definition of the free-stream location. This was achieved by using
continuity to estimate a suitable area variation in the duct, as shown in Fig. 6. A different shape was required for
case T3C4 than the others. Polynomial expressions for the duct shapes are given by Suluksna et al.7 The mesh size
used was 60x146 cells, with a near-wall mesh spacing of 10-5m. Mesh independence will be confirmed for the final
paper.
Since the data indicate elevated levels of turbulence intensity at y=δ99, the free stream edge location was taken to
be y=2δ99 for the purposes of these simulations. Reference turbulence intensities and streamwise velocities at y=2δ99
were linearly interpolated from the data at each measurement station. A polynomial curve fit to the data was used to
specify the free stream location to fulfill the requirement of the γ-Reθ model.
Material properties were the same as for the zero-
pressure gradient cases. Slip walls were used on the Case Uin (m/s) Tu (%) µt/µ
top boundary and upstream of the plate. The inflow T3C1 6.1 7.78 44
boundary conditions are reported in Table 2. To
T3C2 5.25 3.1 9
facilitate a better match to the turbulence intensity
T3C3 3.95 3.1 6
profiles, the turbulence decay was deactivated
T3C4 1.28 2.8 2
upstream of the plate, so that the inlet values are, in
effect, leading edge values. These conditions were T3C5 8.8 4.5 11
specified in terms of turbulence intensity and viscosity
Table 2. Inflow boundary conditions for non-zero-
ratio, Tu = 2 / 3k / U and µt / µ = ρ k / (ωµ ) . pressure gradient flat plate simulations.
Figs. 7 and 8 show the correspondence of the free-
5
American Institute of Aeronautics and Astronautics
092407
 Figure 6. Mesh schematic for non-zero-pressure gradient flat plate simulations.

Figure 7. Free-stream velocity profiles for non- Figure 8. Free-stream turbulence decay for non-
zero-pressure gradient cases. zero-pressure gradient cases.

Figure 9. Skin friction coefficient results for Figure 10. Skin friction coefficient results for cases
cases T3C1, T3C3 and T3C5. T3C2 and T3C4.

stream velocity and turbulence intensity to the data. Some discrepancies occur near the leading edge of the plate (x/L
< 0.2). These reflect the compromises that were required to reasonably match the free stream conditions over the
bulk of the plate.
Skin friction coefficient results for cases T3C1, T3C3 and T3C5 are shown in Fig. 9 and for T3C2 and T3C4 in
Fig 10. For simplicity, the wall shear stress is non-dimensionalized by the inflow velocity rather than the free stream
flow velocity (the experimental data as well as the simulation results).
In general, the results are similar to those reported by Langtry,5 although by using a boundary layer code for
cases T3C2, T3C3 and T3C5, Langtry5 was able to exercise more control over the free-stream conditions. Case
T3C1 (not reported by Langtry5) experiences an early onset of transition due to the high free-stream turbulence
intensities. Similar to the T3B case, skin friction coefficient is somewhat high in the transition region. Case T3C2
represents transition near the suction peak, and the onset is predicted too far downstream, consistent with the results
of Langtry5. The results of case T3C3, where separation occurs in an adverse pressure gradient are excellent. Case
6
American Institute of Aeronautics and Astronautics
092407
T3C4 represents transition due to the presence of a laminar separation bubble, which was properly captured,
although a little far downstream. Finally, the higher Reynolds number case T3C5 shows the behavior of the model in
a favorable pressure gradient. Consistent with the results of Langtry5, the onset of transition is slightly too far
downstream.
One very important outcome of these non-zero-pressure gradient flat plate cases was the conclusion that the
function F in Eq. (24) needs to be set to unity for best results. In other words, the effects of stream-wise pressure
gradient were explicitly omitted in the evaluation of Reθt. Suluksna et al.7 justify this omission with the explanation
that, by expressing Reθt as a function of Tu, pressure gradient effects are implicitly accounted for because of the
effect of the pressure gradient on the local turbulence in the boundary layer. Simply put, adverse pressure gradients
tend to promote turbulence whereas favorable pressure gradients suppress it. This assertion must be considered in
the light of the fact that the original correlation of Eqs. (19)-(23) was based on the leading edge values of Tu, not
local values.

V. Two-Dimensional Test Cases

Three 2D test cases are presented in this section: the Aerospatiale-A airfoil16, the Zierke and Deutsch compressor
cascade17 and the VKI BRITE large scale turbine cascade measured at the University of Genoa. 18 Each of these
cases have been considered by Langtry.5
Two sets of experimental data are available for the Aerospatiale-A airfoil, the F1 wind tunnel data at a 13.1°
angle of attack at a Reynolds number of 2.07x106, and the F2 wind tunnel data at 13.3° angle of attack and at a
Reynolds number of 2.1x106. The former condition was simulated herein, although F2 data for the skin friction and
force coefficients are used as a basis of comparison. The C-topology computational mesh of 200x800 cells is shown
in Fig. 11. The near-wall mesh spacing is 10-5 chord lengths, resulting in a peak value of y+ ranging from
approximately 0.25 on the pressure surface to a peak of 0.7 immediately downstream of transition on the suction
surface. This mesh is fine enough to yield a mesh-independent result. The domain extends from 8 chord lengths
upstream to 20 chord lengths downstream. Since the inflow boundary is non-planar, the inflow values of k and ω
were specified with the appropriate analytical decay laws to obtain Tu = 0.2% and µ/µt = 10 at the leading edge
plane. The free-stream edge definition for the γ-Reθ model was arbitrarily defined as 0.05 chord lengths from the
airfoil surface. The simulation results for pressure coefficient are compared to the F2 experimental data in Fig. 12,
and the suction surface skin friction coefficient is compared to both the F1 and F2 data in Fig 13. The most
significant deviation in the pressure coefficient distribution occurs at the trailing edge. This might be due to the fact
that a sharp trailing edge was modeled. Transition on the suction surface is triggered by a laminar separation bubble
at about 12% of chord. Turbulent separation occurs at about 96% chord, somewhat later than the experiment. The
predicted lift and drag coefficients are 1.445 and 0.0169, which compare reasonably well with the F1 reported
coefficients of 1.562 and 0.0208, respectively.

Figure 11. Mesh used for the Aerospatiale-A airfoil.

7
American Institute of Aeronautics and Astronautics
092407
 Figure 12. Pressure coefficient results for Figure 13. Skin friction coefficient on the
the Aerospatiale-A airfoil. suction surface for Aerospatiale-A airfoil.
The highly loaded compressor cascade case of Zierke and Deutsch17 was simulated for the case of leading edge
incidence angle of -1.5°. The computational mesh, consisting of 89,752 cells is shown in Fig 14. The wall-adjacent
cell centroid height is 4x10-6m, resulting in typical values of y+ ranging from 0.1 to 0.6. The inflow was specified a
distance of 0.3 axial chord lengths upstream. The inflow turbulence boundary conditions were Tu = 0.2% and µ/µt =
2. The free-stream edge definition for the γ-Reθ model was arbitrarily defined as 5mm (approximately 2.3% of the
axial chord length) from the blade surface.

Figure 14. Mesh used for Zierke and Deutsch compressor cascade.

The pressure coefficient distribution is shown in Fig. 15 and the skin friction coefficient in Fig. 16. The various
experimental data for skin friction represent the different methods used to derive this quantity from the measured
velocity profiles. As reported by Langtry,5 transition occurs immediately on the suction surface, triggered by a small
laminar separation bubble at the leading edge. Transition is also triggered by laminar separation on the suction
surface, but closer to mid-chord. The transition location in the present simulation is slightly further upstream than
shown by the experimental data.

8
American Institute of Aeronautics and Astronautics
092407
 Figure 15. Pressure coefficient results for Figure 16. Skin coefficient results for the
the Zierke and Deutsch compressor cascade. Zierke and Deutsch compressor cascade.

The computational mesh for the Genoa turbine cascade is shown in Fig. 17. An O-type mesh is used to surround the
blade and the total cell count is 79,020. The near-wall cell centroid height was nominally 2x10-6m, although the
mesh smoothing algorithm resulted in some scatter about this value. This mesh spacing resulted in a maximum y+
value of approximately 0.65 on the suction surface.

Figure 17. Mesh used for the Genoa turbine cascade.

The relative inlet total pressure is given as 3060 Pa, and, to match the required outlet isentropic Mach number,
an outlet relative static pressure of -1041.97 was specified. The inflow turbulence boundary conditions were Tu =
4% and µ/µt = 80. These large values were required to match the experimental data in the free stream. The
correspondence is shown in Fig. 18 for specific probe points corresponding to the outside edge of the boundary layer
traverse planes. These turbulence conditions were considerably larger than the values used by Langtry,5 who chose
respective values of 1.5% and 1.5. The free-stream edge definition for the γ-Reθ model was arbitrarily defined as
5mm (approximately 1.7% of the blade chord length) from the blade surface.
The isentropic velocity distribution is shown in Fig. 18 and the suction surface skin friction velocity, normalized
by free-stream velocity, is shown in Fig. 19. The pressure surface skin friction is not plotted since the boundary
layer remains laminar on that surface. Transition occurs somewhat further upstream on the suction surface than in
9
American Institute of Aeronautics and Astronautics
092407
the experiment. The results shown by Langtry5 also
reflect an early transition, but not to the same extent.
We attribute the slight discrepancy between our results
and Langtry’s to the choice of inflow turbulence
conditions. Indeed, experimenting with the inflow
turbulence showed that reducing the inflow turbulence
intensity will delay transition, as might be expected.

Figure 18. Free-stream turbulence intensity at

boundary layer edge probe locations for the
Genoa turbine cascade.

Figure 19. Isentropic velocity distribution for the Figure 20. Normalized friction velocity on the
Genoa turbine cascade. Genoa turbine suction surface.

VI. Three-Dimensional Multi-Element Airfoil

The objective of the 3D multi-element airfoil simulation was

to evaluate the performance of the γ-Reθ transition model on
a typical industrial application for which it was known a
priori that transition prediction was important. The case
selected was a Formula One car rear wing configuration.
Wind tunnel flow visualization showed that a laminar
separation occurred on the suction surface of the upper airfoil
flap element, in addition to the turbulent separation further
downstream. Flow simulations without a transition model
can, therefore, not be relied on to accurately predict the force
coefficients. The geometry and symmetry plane mesh is
shown in Fig. 21, with a close-up of the surface mesh detail
in Fig 22. The mesh comprised 12.8 million cells, and was
created within STAR-CCM+ using the trimmed mesh
approach. This approach uses a Cartesian mesh with hanging Figure 21 Multi-element airfoil geometry and
node refinement in the far field and achieves adequate symmetry plane mesh.
10
American Institute of Aeronautics and Astronautics
092407
 boundary layer resolution by extruding a prism mesh layer
in a void created in the vicinity of the surfaces. The axial
chord of the upper two-element airfoil is 0.334m at the
symmetry plane. The Reynolds number based on this
chord was 1.1x106. The wind tunnel inlet was placed
approximately 9m upstream of the upper airfoil leading
edge. Slip walls were used on all wind tunnel boundaries
to avoid the need to resolve the boundary layers. The
specified inflow turbulence boundary conditions were Tu =
2% and µ/µt = 100, resulting in a free-stream turbulence
intensity at the airfoil leading edge of 0.45%. The free-
stream edge definition for the γ-Reθ model was arbitrarily
defined as 3mm (approximately 1% of the axial chord)
from the airfoil surface.
In industrial situations, mesh resolution is often driven
by available computational resources, rather than the Figure 22. Mesh detail on airfoil surface and
proper approach of demonstrating mesh independence. The symmetry plan.
current mesh is a representative example of the degree of
resolution that a sophisticated industrial CFD practitioners might use. Nevertheless, a 25 million cell version will be
run for the final paper to investigate the effects of further mesh refinement.
Fig. 23 shows the wind tunnel oil streak flow visualization on the suction surface of the flap. A similar view of
streakline visualization from the STARCCM+ simulation is shown in Fig. 24. It is clear that the major features of
laminar separation, transition to turbulence, and turbulent separation have been captured in the simulation, although
the spanwise distribution of the turbulent separation location is not perfectly replicated. Still, this example shows
that the γ-Reθ transition model can produce realistic flow features on a realistic 3D geometry and a reasonably
representative mesh. Pressure coefficient data will also be shown in the final paper.

Figure 23. Oil streak flow visualization on the Figure 24. Streakline visualization colored by
pressure surface of airfoil flap. skin friction coefficient on pressure surface of
airfoil flap.

VII. Conclusion
The γ-Reθ transition model has been successfully implemented in STAR-CCM+, a commercial unstructured
CFD code. After a process of calibration using published experimental data, the proprietary correlations omitted
from publications by the originators of the model were synthesized. Sufficient information is included in this paper
to guide others to perform a similar calibration. Using the synthesized correlations, the model was applied to several
validation cases. The results of these validation cases compare favorably to the results shown by Langtry5 for the
same test cases.

11
American Institute of Aeronautics and Astronautics
092407
 During the course of this study, it was observed that the computational costs are significantly higher than fully
turbulent calculations for two reasons. First, requirements on mesh resolution are greater (both wall-normal and
streamwise spacing). Second, apart from the overhead of solving two additional transport equations, the interaction
between the momentum, turbulence and transition equations requires more iterations for convergence. In addition,
much more attention needs to be paid to the inflow or free-stream turbulence boundary conditions. Nevertheless, the
successful application of the model to a realistic industrial flow simulation in this study, the Formula One multi-
element rear wing, illustrates the value of the approach.

Appendix
11,12
The SST k-ω turbulence model, as modified for use with the γ-Reθ transition model1-5, consists of two
transport equations, for the turbulent kinetic energy, k, and the specific dissipation rate, ω,
Dk ∂  ∂k   ∂u j 
ρ = ( µ + σ k µt )
Dt ∂x j 
 +  γ eff µt S − ρ k
∂x j  
2

∂x j 
( )
 − min  max γ eff , 0.1 ,1 ρβ ω k ,
*
(3)

Dω ∂  ∂ω   ∂u j  1 δ k δω
( µ + σ ω µt )  + α  µt S − ρ k  − ρβω + 2 ρ (1 − F1 )σ ω 2
2 2
ρ = , (4)
Dt ∂x j  ∂x j  ∂x  ω δ xj δ xj
 j 

where S is the strain rate tensor modulus and γeff is the effective intermittency.
The turbulent viscosity is defined as:
 1 0.6 
µt = ρ kT , T = min  , , (5a, b)
 max [ω , SF2 / a1 ] 3S 

where ρ is the density, uj is the velocity vector, µ is the molecular viscosity, and µt is the eddy viscosity. The
realizability constraint of Durbin13 is reflected in the time scale (5a).
The functions F1 and F2 are given by:
  k 500ν  2k   1 δ k δω 
F1 = max  F3 , tanh arg14  , arg1 = max min 
( ) ,  , 2  , CDkω = max  ,
    0.09ω y ω y  y CD 
2  ω δ xj δ xj 
  kω   
(6a, b, c)
 2 k 500ν 
( )
F2 = tanh arg 22 , arg 2 = max 
 0.09ω y ω y 2 
, , (7a, b)
 
( )
F3 = exp −( R y / 120)8 , R y = ρ yk 1/ 2 / µ , (8a, b)
where y is the normal distance from the nearest wall.
The coefficients φ of the model are calculated from the blending function

φ = F1φ1 + (1 − F1 )φ2 , (9)

where the coefficients φ1, φ2 are:

σ k1 = 0.85 , σ ω1 = 0.65 , β1 = 0.075 , α1 = β1 / β * − σ ω1κ 2 / β * ,

σ k 2 = 1 .0, σ ω 2 = 0.856 , β 2 = 0.0828 , α 2 = β 2 / β * − σ ω 2κ 2 / β * ,

and a1 = 0.3, κ = 0.41, β* = 0.09.

A Neumann (zero-flux) wall boundary condition is specified for k, whereas ω = 6ν/(βy2) is specifically set in the
wall-adjacent cells.

12
American Institute of Aeronautics and Astronautics
092407
 The γ-Reθ transition model1-5 consists of two transport equations: an equation for the intermittency, γ, and an
equation for the transported transition momentum thickness Reynolds number, Reθt. These transport equations are
written as

Dγ ∂  µ  ∂γ 
ρ =  µ + t   + F c ρ S (γ Fonset )0.5 (1 − ce1γ ) + ca 2 ρΩγ Fturb (1 − ce 2γ ) , (10)
Dt ∂x j 
 σγ  ∂x j  length a1
 


D Reθ t ∂ 

∂ Re ( ρU ) 2
ρ = σ θ t ( µ + µt )
θt
 + cθ t
)(1 − F ) ,
(Reθ t − Re (11)
θt θt
Dt ∂x j  ∂x j  500µ

where Ω is the vorticity tensor modulus, and U is the local velocity magnitude. The parameters Flength and Fonset are
used to control the length and onset location of transition respectively. Fturb and Fθt are the parameters for controlling
the destruction/relaminarization of the boundary layer and the boundary layer detector respectively.
The modeled transport equations are controlled by the following functions:
Fonset = max( Fonset 2 − Fonset 3 , 0) , Fturb = exp  − ( RT / 4 )  ,
4

(12a, b)
 
Reν
Fonset 2 = min  max( Fonset1 , Fonset
4

1 ), 2  , Fonset1 = , (13a, b)
2.193Reθ c

(
Fonset 3 = max 1 − ( RT / 2.5) , 0 ,
3
) (14)

  
    c γ −1   
4 2
 U2    Re 2 
Fθ t = min max  Fwake ⋅ exp  −    ,1 −  e 2 
 ,1 , Fwake = exp  −  5   ,
ω
(15a, b)
   375Ων Re
  ce 2 − 1      10  
   θt     
ρ Sy 2 ρk ρω y 2
Reν = , RT = , Reω = , (16a, b, c)
µ µω µ
where Reν is the strain rate Reynolds number, RT is a turbulent Reynolds number (commonly termed the viscosity
ratio. Reθc is the critical momentum thickness Reynolds number where the intermittency first appears in the
boundary layer.
The model constants are ca1=2.0, ce1=1.0, ca2=0.06, ce2=50.0, cα=0.5, σγ =1.0, σθt=2.0, cθt=0.03.
Neumann (zero-flux) wall boundary conditions are applied for γ and Reθt. At inlets, γ=1.0 is applied and Reθt is
obtained from the freestream correlation for Reθt.
The effective intermittency, γeff, is obtained from

γ eff = max(γ , γ sep ) , (17)

   
Reν
( )
4

γ sep = min  2 max  − 1, 0  Freattach , 2 Fθ t , Freattach = exp − ( RT / 20 ) . (18a, b)

  3.235Reθ c  

To close the γ-Reθ transition model, three correlations are required: for Reθt, Reθc and Flength. Two correlations
have been proposed for Reθt, the transition onset momentum thickness Reynolds number, defined in the free stream,
based on a range of experimental data. The expression proposed Menter et al. 1 is:
− 1.027
Reθ t = 803.73 (Tu + 0.6067 ) Fλ , K , (19)
1 − Fλ ⋅ e − Tu /3
;λ ≤ 0
Fλ , K =  (20)
(
1 + FK ⋅ 1 − e
−2Tu /3
) (
+ 0.556 1 − e −23.9λ ⋅ e −Tu /3 ) ;λ > 0
Fλ = −10.32λ − 89.47λ 2 − 265.51λ 3 , (21a)
2 3
(
FK = 0.0962 K ⋅10 + 0.148 K ⋅10 6
) ( 6
) + 0.0141 K ⋅ 10( 6
), (22b)

13
American Institute of Aeronautics and Astronautics
092407
 θ 2 dU ν dU 100(2k / 3)1/ 2 dU ∂u j ui u j
λθ = ⋅ ,K = 2 ⋅ , Tu = , = . (23a, b, c, d)
ν ds U ds U ds ∂xi U 2

The correlation of Eq. (19) was later modified by Langtry5 to the following form:
 0.2196 
 1173.51 − 589.428Tu + F ( λ , Tu ) ; Tu ≤ 1.3
Reθ t =   Tu 2  (24)
331.5 Tu − 0.5658 −0.671 F λ , Tu ;
 [ ] ( ) Tu > 1.3
1 + e − (2Tu /3) ⋅ 12.986λ + 123.66λ 2 + 405.689λ 3
( )
1.5

 ;λ ≤ 0
F (λ , Tu ) =  (25)
(
−2Tu
1 + 0.275e ⋅ 1 − e )
( −35 λ )
;λ > 0

Acknowledgments
The authors are grateful to Dr. Gary Ahlin of Computational Dynamics, London for providing the multi-element
airfoil mesh and to Pointwise Inc. for the use of their Gridgen software used to create all the 2D meshes. Paul Malan
thanks Professor Erik Dick of the University of Ghent and Professor Witold Elsner of Czetochowa University of
Technology for useful and frank discussion during the initial stages of this work.

References
1
Menter, F.R., Langtry, R.B., Likki, S.R., Suzen, Y.B., Huang, P.G., and Völker, S., “A Correlation-based Transition Model
Using Local Variables Part 1 – Model Formulation,” ASME GT2004-53452, Proceedings of the ASME Turbo Expo, Power for
Land Sea and Air, June 14-17, 2004.
2
Langtry, R.B., and Menter, F.R., “Transition Modeling for General CFD Applications in Aeronautics”, AIAA Paper 2005-
522, Reno, Nevada, 2005.
3
Menter, F.R., Langtry, R.B., Völker, S.and Huang, P.G., “Transition Modelling for General Purpose CFD Codes,”
ERCOFTAC Int. Symp. Engineering Turbulence Modelling and Measurements, ETMM6, 2005.
4
Menter, F.R., Langtry, R., and Volker, S., “Transition Modelling for General Purpose CFD Codes,” Flow, Turbulence and
Combusion, Vol. 77, pp.277-303, 2006.
5
Langtry, R.B., “A Correlation-based Transition Model Using Local Variables for Unstructured Parallelized CFD Codes,”
Dr.-Ing thesis, Institute of Thermal Turbomachinery and Machinery Laboratory, University of Stuttgart, 2006.
6
Piotrowski., W., Elsner, W. and Drobniak, S., “Transition Prediction on Turbine Blade Profile with Intermittency Transport
Equation,” ASME GT2008-50796, Proceedings of the ASME Turbo Expo, Power for Land Sea and Air, June 9-13, 2008.
7
Suluksna, K., Dechaumphai, P. and Juntasaro E., “Correlations for Modeling Transitional Boundary Layers under Influences
of Freestream Turbulence and Pressure Gradient,” Int. J. Heat and Fluid Flow (to be published).
8
STAR-CCM+ Software Package, Version 3.06, CD-adapco, Melville NY (to be released).
9
Demirdzic, I. and Muzaferija, S., “Numerical Method for Coupled Fluid Flow, Heat Transfer and Stress Analysis Using
Unstructured Moving Meshes with Cells of Arbitrary Topology,” Comput. Methods Appl. Mech. Engrg., 1995.
10
Weiss, J.M. and Smith, W.A., “Preconditioning Applied to Variable and Constant Density Flows,” AIAA Journal, Vol. 33,
No. 11, 1995, pp. 2050-2057.
11
Menter, F. R., “Two-Equation Eddy Viscosity Models for Engineering Applications,” AIAA Journal, Vol. 32, No. 8, 1994,
pp. 1598-1605.
12
Vieser, W., Thomas Esch, T. and Menter, F., “Heat Transfer Predictions using Advanced Two-Equation Turbulence
Models, ” CFX Technical Memorandum CFX-VAL10/0602, ANSYS Inc., 2002.
13
Durbin, P., “On the k-ε Stagnation Point Anomaly,” International Journal of Heat and Fluid Flow, Vol. 17, pp. 89-90,
1995.
14
ERCOFTAC (European Research Community on Flow, Turbulence, Combustion) Nexus. [online database], URL:
http://ercoftac.mech.surrey.ac.uk/ [cited 15 Feb 2008].
15
Schubauer, G.B. and Klebanoff, P.S. “Contributions on the Mechanics of Boundary Layer Transition,” NACA Technical
Note 3489, 1955.
16
Chaput, E., “Chapter III: Application-Oriented Synthesis of Work Presented in Chapter II,” Notes on Numerical Fluid
Mechanics, Vol. 58, Vieweg Braunschweig, Wiesbaden, pp. 327-346, 1997.
17
Zierke, W.C. and Deutsch, S., “The Measurement of Boundary Layers on a Compressor Blade in Cascade, Vol. 1 ---
Experimental Technique, Analysis and Results,” NASA CR 185118, 1989.
18
Ubaldi, M., Zunino, P., Campora, U. and Ghiglione, A., “Detailed Velocity and Turbulence Measurements of the Profile
Boundary Layer in a Large Scale Turbine Cascade,” Proceedings of the International Gas Turbine and Aeroengine Congress and
Exhibition, Birmingham, UK, ASME 96-GT-42, 1996.
14
American Institute of Aeronautics and Astronautics
092407